RUBIDIUM, A Program for Computer- Aided Assignment of Two

RUBIDIUM, A Program for Computer- Aided Assignment of Two-Dimensional NMR Spectra of. Polypeptides. CHIN YU,*vt JAN-FU HWANG; TUNG-BO CHEN,* and VON-W...
17 downloads 9 Views 528KB Size
183

J. Chem. Zt$ Comput. Sci. 1992, 32, 183-187

RUBIDIUM,

A Program for Computer- Aided Assignment of Two-Dimensional NMR Spectra of Polypeptides CHIN YU,*vt JAN-FU HWANG; TUNG-BO CHEN,* and VON-WUN SOOt

Department of Chemistry and Department of Computer Science, National Tsing Hua University, Hsinchu, Taiwan 30043, Republic of China Received October 7, 1991 Taking advantage of the rule-based expert system technology, a program named RUBIDIUM (Rule-Based Identification In 2D NMR Spectrum) was developed to accomplish the automatic 'H NMR resonance assignments of polypeptides. Besides noise elimination and peak selection capabilities, RUBIDIUM detects the cross-peak patterns of amino acid residues in the COSY spectrum, assigning these patterns to amino acid types, performing sequential assignments using combined COSY/NOESY spectra, and finally, achieving the total assignment of the 'HNMR spectrum.

INTRODUCTION For a small protein, 2D NMR' is a powerful technique to determine the 3D structure in a solution state, whereas the X-ray technique is applicable for the single crystal. Whenever proteins cannot be crystallized, or when the behavior of a protein in solution is to be studied, NMR spectroscopy becomes indispensible. The NMR determination of the 3D structure of a protein typically passes three stages: (a) acquisition of 2D NMR data; (b) peak assignment of the spectrum resonances, and consequent deduction of distance constraints; and (c) generation of a molecular structure which satisfies the constraints. Among them, the peak assignment process in stage b is highly complex and time-consuming;several months may be required to accomplish it. It is obvious that computer programs to alleviate this task would be useful. Several methodologies appear promising,2* but we decided to take advantage of the rule-based expert system to accomplish the 'HNMR resonance automatic assignment for the polypeptides. In this paper, we present the program that we developed, named RUBIDIUM (Rule-Based Identification In 2D NMR Spectrum). Two polypeptide samples, oxytocin and vasopressin, were tested. MATERIALS AND METHODS Vasopressin and oxytocin were obtained commercially (Sigma). Each peptide (10 mg) was dissolved in a DMSO-d, (Aldrich) solution (500 p L ) to make the concentration -20 1-sulfonate was used mmol L-' . 4,4-Dimethyl-4-silapentaneas an internal standard. The NMR tube was degassed and sealed. The NMR experiments were executed on a 400-MHz spectrometer (Bruker AM-400). The doublequantum filtered COSY'O (DQF-COSY) experiment was carried out in the phasesensitive mode to obtain J-connectivities. The 2D NOE (NOESY)"-13 experiment was also performed in the phasesensitive mode with a mixing period of 120 ms. Data Preprocessing. 2D NMR data were acquired in Bruker Aspect-3OOO and transferred to a pVaxIII (MV 3600) computer. The data were processed with a FTNMR program (Hare Research) in pVaxIII to generate a SMX file and then transferred to a SUN 386i computer for noise elimination and peak selection. To whom correspondence should bc addressed. t Department of Chemistry.

'kpartment of Computer Science.

Noise Elimination. The 2D data matrix was divided into 16 areas along the x-axis (w2 dimension). The average value was calculated for each column and set to be the threshold. Any point along the column with a value smaller the threshold is set to 0. The tl noise is further eliminated according to the approach expressed in the following two equations: If llZil - lZjll 4 min(Zi,Zj), then Zi = Zj = (Zi If llZil - lZjll

+ Zj)/2

(1)

> min(Zi,Zj), then Zi = Z, = min(Zi,Zj) (2)

Thus for any point i with intensity Zi (positive or negative value) in the 2D spectrum, there is a symmetrical point j (with respect to the diagonal axis) with intensity Zj. The intensities of these two points i and j were set to be the lesser of i and j if the difference between lZil and lZjl was greater than Zi or Zj (eq 2). Otherwise, the intensities of these two points were set to be the average value of Zi and 4 (eq 1). Because the signals in the upper left and lower right regions along the diagonal axis in 2D spectra should be symmetric,14the asymmetric signals were regarded as noise and thus eliminated. After this process, the unsymmetrical cross peaks (with respect to the diagonal) were suppressed, and the cross peaks masked by the t, noise were enhanced. Peak Selection. After noise elimination, a peak was listed if its intensity exceeded the threshold. For the NOESY spectrum, the selected peaks were directly listed. For DQFCOSY, the peak was not listed until there was a minimum of two-thirds of the anti-phase fine structurelo component. Figure 1 shows the COSY spectrum after peak selection was done. The Main Program: RUBIDIUM. RUBIDIUM is written in CLIPS15J6and C languages. It can be run on any computer that supports the C language. It is important to have enough memory (RAM)for the computer to run RUBIDIUM. On the personal computer (IBM PC-type), for example, if there are 80 cross peaks for COSY and 150 for NOESY spectra, RUBIDIUM requires 640K RAM (including CLIPS interpreter). Other information about RUBIDIUM are the following: Data Input. To execute RUBIDIUM, the following data input are needed: (a) amino acid sequence of peptide (b) table of cross peaks from the COSY spectrum (c) table of cross peaks from the NOESY spectrum

0095-2338/92/1632-0183$03.00/00 1992 American Chemical Society

184 J. Chem. If. Comput. Sci., Vol. 32, No. 3, 1992

d

YU ET

AL.

a

Figure 1. Result of oxytocin COSY spectrum after peak selection. Because the spectrum is symmetric about the diagonal axis, only the lower right region is shown.

The input format for the amino acid sequence of oxytocin, for example, should be typed as follows: CYS TYR ILE GLN ASN CYS PRO LEU GLY in which the three-letter code represents the amino acid from the N-terminus (CYS) on the left to the C-terminus (GLY) on the right. Each amino acid is represented by a three-letter code. Pattern Recognition. Each amino acid residue has a different chemical shift in the ‘H NMR spectrum; this effect causes the cross peak in the COSY spectrum to have a specific pattern. From each cross-peak pattern in the COSY spectrum, the amino acid residue was recognized. The cross-peak pattern of 20 amino acid residues were generally classified into three categories:18 (1) Unique spin system; amino acid residues include Gly, Ala, Val, Ile, Leu, and Thr; the cross-peak pattern for these amino acid residues is unique (2) Threespin system; in this category, there are two CSH protons that are not degenerated and shifted relatively upfield; the amino acids such as Asn, Asp, Cys, Ser, Phe, Tyr, His, and Trp belong to this class (3) Long side-chain system; the amino acids with a long side chain such as Glu, Gln, Met, Pro, Arg, and Lys belong to this family. The NOESY spectrum was also taken into consideration by RUBIDIUM to enhance the accuracy for pattern recognition. This effect is important for Gln, Glu, Phe, and Tyr; for these amino acids, there are NOESY cross peaks for (CQ, C6H) and ( O H , C‘H). Pro is the only amino acid that has

’’

no amide proton; however, this case can be recognized by the (NH, C”H) cross peak in the NOESY spectrum. The pattern-matching rule for the amino acid glycine (Gly) to be recognized, for example, is written as a CLIPS16 rule: (defrule m a r k G L Y (exist GLY ?n) (peak cosy ?NH&: (< (abs(-?NH 8.1)) 1) ?aH&: (< (abs(- ?aH 3.8)) 1) ?x ?y) (peak cosy ?N1&: (